fs/ceph/super.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2
   3 #include <linux/ceph/ceph_debug.h>
   4
   5 #include <linux/backing-dev.h>
   6 #include <linux/ctype.h>
   7 #include <linux/fs.h>
   8 #include <linux/inet.h>
   9 #include <linux/in6.h>
  10 #include <linux/module.h>
  11 #include <linux/mount.h>
  12 #include <linux/fs_context.h>
  13 #include <linux/fs_parser.h>
  14 #include <linux/sched.h>
  15 #include <linux/seq_file.h>
  16 #include <linux/slab.h>
  17 #include <linux/statfs.h>
  18 #include <linux/string.h>
  19
  20 #include "super.h"
  21 #include "mds_client.h"
  22 #include "cache.h"
  23
  24 #include <linux/ceph/ceph_features.h>
  25 #include <linux/ceph/decode.h>
  26 #include <linux/ceph/mon_client.h>
  27 #include <linux/ceph/auth.h>
  28 #include <linux/ceph/debugfs.h>
  29
  30 #include <uapi/linux/magic.h>
  31
  32 static DEFINE_SPINLOCK(ceph_fsc_lock);
  33 static LIST_HEAD(ceph_fsc_list);
  34
  35 /*
  36  * Ceph superblock operations
  37  *
  38  * Handle the basics of mounting, unmounting.
  39  */
  40
  41 /*
  42  * super ops
  43  */
  44 static void ceph_put_super(struct super_block *s)
  45 {
  46         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
  47
  48         dout("put_super\n");
  49         ceph_mdsc_close_sessions(fsc->mdsc);
  50 }
  51
  52 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
  53 {
  54         struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry));
  55         struct ceph_mon_client *monc = &fsc->client->monc;
  56         struct ceph_statfs st;
  57         int i, err;
  58         u64 data_pool;
  59
  60         if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
  61                 data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
  62         } else {
  63                 data_pool = CEPH_NOPOOL;
  64         }
  65
  66         dout("statfs\n");
  67         err = ceph_monc_do_statfs(monc, data_pool, &st);
  68         if (err < 0)
  69                 return err;
  70
  71         /* fill in kstatfs */
  72         buf->f_type = CEPH_SUPER_MAGIC;  /* ?? */
  73
  74         /*
  75          * Express utilization in terms of large blocks to avoid
  76          * overflow on 32-bit machines.
  77          */
  78         buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
  79
  80         /*
  81          * By default use root quota for stats; fallback to overall filesystem
  82          * usage if using 'noquotadf' mount option or if the root dir doesn't
  83          * have max_bytes quota set.
  84          */
  85         if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
  86             !ceph_quota_update_statfs(fsc, buf)) {
  87                 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
  88                 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  89                 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
  90         }
  91
  92         /*
  93          * NOTE: for the time being, we make bsize == frsize to humor
  94          * not-yet-ancient versions of glibc that are broken.
  95          * Someday, we will probably want to report a real block
  96          * size...  whatever that may mean for a network file system!
  97          */
  98         buf->f_bsize = buf->f_frsize;
  99
 100         buf->f_files = le64_to_cpu(st.num_objects);
 101         buf->f_ffree = -1;
 102         buf->f_namelen = NAME_MAX;
 103
 104         /* Must convert the fsid, for consistent values across arches */
 105         buf->f_fsid.val[0] = 0;
 106         mutex_lock(&monc->mutex);
 107         for (i = 0 ; i < sizeof(monc->monmap->fsid) / sizeof(__le32) ; ++i)
 108                 buf->f_fsid.val[0] ^= le32_to_cpu(((__le32 *)&monc->monmap->fsid)[i]);
 109         mutex_unlock(&monc->mutex);
 110
 111         /* fold the fs_cluster_id into the upper bits */
 112         buf->f_fsid.val[1] = monc->fs_cluster_id;
 113
 114         return 0;
 115 }
 116
 117 static int ceph_sync_fs(struct super_block *sb, int wait)
 118 {
 119         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 120
 121         if (!wait) {
 122                 dout("sync_fs (non-blocking)\n");
 123                 ceph_flush_dirty_caps(fsc->mdsc);
 124                 dout("sync_fs (non-blocking) done\n");
 125                 return 0;
 126         }
 127
 128         dout("sync_fs (blocking)\n");
 129         ceph_osdc_sync(&fsc->client->osdc);
 130         ceph_mdsc_sync(fsc->mdsc);
 131         dout("sync_fs (blocking) done\n");
 132         return 0;
 133 }
 134
 135 /*
 136  * mount options
 137  */
 138 enum {
 139         Opt_wsize,
 140         Opt_rsize,
 141         Opt_rasize,
 142         Opt_caps_wanted_delay_min,
 143         Opt_caps_wanted_delay_max,
 144         Opt_caps_max,
 145         Opt_readdir_max_entries,
 146         Opt_readdir_max_bytes,
 147         Opt_congestion_kb,
 148         /* int args above */
 149         Opt_snapdirname,
 150         Opt_mds_namespace,
 151         Opt_recover_session,
 152         Opt_source,
 153         Opt_mon_addr,
 154         /* string args above */
 155         Opt_dirstat,
 156         Opt_rbytes,
 157         Opt_asyncreaddir,
 158         Opt_dcache,
 159         Opt_ino32,
 160         Opt_fscache,
 161         Opt_poolperm,
 162         Opt_require_active_mds,
 163         Opt_acl,
 164         Opt_quotadf,
 165         Opt_copyfrom,
 166         Opt_wsync,
 167         Opt_pagecache,
 168 };
 169
 170 enum ceph_recover_session_mode {
 171         ceph_recover_session_no,
 172         ceph_recover_session_clean
 173 };
 174
 175 static const struct constant_table ceph_param_recover[] = {
 176         { "no",         ceph_recover_session_no },
 177         { "clean",      ceph_recover_session_clean },
 178         {}
 179 };
 180
 181 static const struct fs_parameter_spec ceph_mount_parameters[] = {
 182         fsparam_flag_no ("acl",                         Opt_acl),
 183         fsparam_flag_no ("asyncreaddir",                Opt_asyncreaddir),
 184         fsparam_s32     ("caps_max",                    Opt_caps_max),
 185         fsparam_u32     ("caps_wanted_delay_max",       Opt_caps_wanted_delay_max),
 186         fsparam_u32     ("caps_wanted_delay_min",       Opt_caps_wanted_delay_min),
 187         fsparam_u32     ("write_congestion_kb",         Opt_congestion_kb),
 188         fsparam_flag_no ("copyfrom",                    Opt_copyfrom),
 189         fsparam_flag_no ("dcache",                      Opt_dcache),
 190         fsparam_flag_no ("dirstat",                     Opt_dirstat),
 191         fsparam_flag_no ("fsc",                         Opt_fscache), // fsc|nofsc
 192         fsparam_string  ("fsc",                         Opt_fscache), // fsc=...
 193         fsparam_flag_no ("ino32",                       Opt_ino32),
 194         fsparam_string  ("mds_namespace",               Opt_mds_namespace),
 195         fsparam_flag_no ("poolperm",                    Opt_poolperm),
 196         fsparam_flag_no ("quotadf",                     Opt_quotadf),
 197         fsparam_u32     ("rasize",                      Opt_rasize),
 198         fsparam_flag_no ("rbytes",                      Opt_rbytes),
 199         fsparam_u32     ("readdir_max_bytes",           Opt_readdir_max_bytes),
 200         fsparam_u32     ("readdir_max_entries",         Opt_readdir_max_entries),
 201         fsparam_enum    ("recover_session",             Opt_recover_session, ceph_param_recover),
 202         fsparam_flag_no ("require_active_mds",          Opt_require_active_mds),
 203         fsparam_u32     ("rsize",                       Opt_rsize),
 204         fsparam_string  ("snapdirname",                 Opt_snapdirname),
 205         fsparam_string  ("source",                      Opt_source),
 206         fsparam_string  ("mon_addr",                    Opt_mon_addr),
 207         fsparam_u32     ("wsize",                       Opt_wsize),
 208         fsparam_flag_no ("wsync",                       Opt_wsync),
 209         fsparam_flag_no ("pagecache",                   Opt_pagecache),
 210         {}
 211 };
 212
 213 struct ceph_parse_opts_ctx {
 214         struct ceph_options             *copts;
 215         struct ceph_mount_options       *opts;
 216 };
 217
 218 /*
 219  * Remove adjacent slashes and then the trailing slash, unless it is
 220  * the only remaining character.
 221  *
 222  * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
 223  */
 224 static void canonicalize_path(char *path)
 225 {
 226         int i, j = 0;
 227
 228         for (i = 0; path[i] != '\0'; i++) {
 229                 if (path[i] != '/' || j < 1 || path[j - 1] != '/')
 230                         path[j++] = path[i];
 231         }
 232
 233         if (j > 1 && path[j - 1] == '/')
 234                 j--;
 235         path[j] = '\0';
 236 }
 237
 238 /*
 239  * Check if the mds namespace in ceph_mount_options matches
 240  * the passed in namespace string. First time match (when
 241  * ->mds_namespace is NULL) is treated specially, since
 242  * ->mds_namespace needs to be initialized by the caller.
 243  */
 244 static int namespace_equals(struct ceph_mount_options *fsopt,
 245                             const char *namespace, size_t len)
 246 {
 247         return !(fsopt->mds_namespace &&
 248                  (strlen(fsopt->mds_namespace) != len ||
 249                   strncmp(fsopt->mds_namespace, namespace, len)));
 250 }
 251
 252 static int ceph_parse_old_source(const char *dev_name, const char *dev_name_end,
 253                                  struct fs_context *fc)
 254 {
 255         int r;
 256         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 257         struct ceph_mount_options *fsopt = pctx->opts;
 258
 259         if (*dev_name_end != ':')
 260                 return invalfc(fc, "separator ':' missing in source");
 261
 262         r = ceph_parse_mon_ips(dev_name, dev_name_end - dev_name,
 263                                pctx->copts, fc->log.log, ',');
 264         if (r)
 265                 return r;
 266
 267         fsopt->new_dev_syntax = false;
 268         return 0;
 269 }
 270
 271 static int ceph_parse_new_source(const char *dev_name, const char *dev_name_end,
 272                                  struct fs_context *fc)
 273 {
 274         size_t len;
 275         struct ceph_fsid fsid;
 276         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 277         struct ceph_mount_options *fsopt = pctx->opts;
 278         char *fsid_start, *fs_name_start;
 279
 280         if (*dev_name_end != '=') {
 281                 dout("separator '=' missing in source");
 282                 return -EINVAL;
 283         }
 284
 285         fsid_start = strchr(dev_name, '@');
 286         if (!fsid_start)
 287                 return invalfc(fc, "missing cluster fsid");
 288         ++fsid_start; /* start of cluster fsid */
 289
 290         fs_name_start = strchr(fsid_start, '.');
 291         if (!fs_name_start)
 292                 return invalfc(fc, "missing file system name");
 293
 294         if (ceph_parse_fsid(fsid_start, &fsid))
 295                 return invalfc(fc, "Invalid FSID");
 296
 297         ++fs_name_start; /* start of file system name */
 298         len = dev_name_end - fs_name_start;
 299
 300         if (!namespace_equals(fsopt, fs_name_start, len))
 301                 return invalfc(fc, "Mismatching mds_namespace");
 302         kfree(fsopt->mds_namespace);
 303         fsopt->mds_namespace = kstrndup(fs_name_start, len, GFP_KERNEL);
 304         if (!fsopt->mds_namespace)
 305                 return -ENOMEM;
 306         dout("file system (mds namespace) '%s'\n", fsopt->mds_namespace);
 307
 308         fsopt->new_dev_syntax = true;
 309         return 0;
 310 }
 311
 312 /*
 313  * Parse the source parameter for new device format. Distinguish the device
 314  * spec from the path. Try parsing new device format and fallback to old
 315  * format if needed.
 316  *
 317  * New device syntax will looks like:
 318  *     <device_spec>=/<path>
 319  * where
 320  *     <device_spec> is name@fsid.fsname
 321  *     <path> is optional, but if present must begin with '/'
 322  * (monitor addresses are passed via mount option)
 323  *
 324  * Old device syntax is:
 325  *     <server_spec>[,<server_spec>...]:[<path>]
 326  * where
 327  *     <server_spec> is <ip>[:<port>]
 328  *     <path> is optional, but if present must begin with '/'
 329  */
 330 static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
 331 {
 332         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 333         struct ceph_mount_options *fsopt = pctx->opts;
 334         char *dev_name = param->string, *dev_name_end;
 335         int ret;
 336
 337         dout("%s '%s'\n", __func__, dev_name);
 338         if (!dev_name || !*dev_name)
 339                 return invalfc(fc, "Empty source");
 340
 341         dev_name_end = strchr(dev_name, '/');
 342         if (dev_name_end) {
 343                 /*
 344                  * The server_path will include the whole chars from userland
 345                  * including the leading '/'.
 346                  */
 347                 kfree(fsopt->server_path);
 348                 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
 349                 if (!fsopt->server_path)
 350                         return -ENOMEM;
 351
 352                 canonicalize_path(fsopt->server_path);
 353         } else {
 354                 dev_name_end = dev_name + strlen(dev_name);
 355         }
 356
 357         dev_name_end--;         /* back up to separator */
 358         if (dev_name_end < dev_name)
 359                 return invalfc(fc, "Path missing in source");
 360
 361         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
 362         if (fsopt->server_path)
 363                 dout("server path '%s'\n", fsopt->server_path);
 364
 365         dout("trying new device syntax");
 366         ret = ceph_parse_new_source(dev_name, dev_name_end, fc);
 367         if (ret) {
 368                 if (ret != -EINVAL)
 369                         return ret;
 370                 dout("trying old device syntax");
 371                 ret = ceph_parse_old_source(dev_name, dev_name_end, fc);
 372                 if (ret)
 373                         return ret;
 374         }
 375
 376         fc->source = param->string;
 377         param->string = NULL;
 378         return 0;
 379 }
 380
 381 static int ceph_parse_mon_addr(struct fs_parameter *param,
 382                                struct fs_context *fc)
 383 {
 384         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 385         struct ceph_mount_options *fsopt = pctx->opts;
 386
 387         kfree(fsopt->mon_addr);
 388         fsopt->mon_addr = param->string;
 389         param->string = NULL;
 390
 391         return ceph_parse_mon_ips(fsopt->mon_addr, strlen(fsopt->mon_addr),
 392                                   pctx->copts, fc->log.log, '/');
 393 }
 394
 395 static int ceph_parse_mount_param(struct fs_context *fc,
 396                                   struct fs_parameter *param)
 397 {
 398         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
 399         struct ceph_mount_options *fsopt = pctx->opts;
 400         struct fs_parse_result result;
 401         unsigned int mode;
 402         int token, ret;
 403
 404         ret = ceph_parse_param(param, pctx->copts, fc->log.log);
 405         if (ret != -ENOPARAM)
 406                 return ret;
 407
 408         token = fs_parse(fc, ceph_mount_parameters, param, &result);
 409         dout("%s fs_parse '%s' token %d\n", __func__, param->key, token);
 410         if (token < 0)
 411                 return token;
 412
 413         switch (token) {
 414         case Opt_snapdirname:
 415                 kfree(fsopt->snapdir_name);
 416                 fsopt->snapdir_name = param->string;
 417                 param->string = NULL;
 418                 break;
 419         case Opt_mds_namespace:
 420                 if (!namespace_equals(fsopt, param->string, strlen(param->string)))
 421                         return invalfc(fc, "Mismatching mds_namespace");
 422                 kfree(fsopt->mds_namespace);
 423                 fsopt->mds_namespace = param->string;
 424                 param->string = NULL;
 425                 break;
 426         case Opt_recover_session:
 427                 mode = result.uint_32;
 428                 if (mode == ceph_recover_session_no)
 429                         fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
 430                 else if (mode == ceph_recover_session_clean)
 431                         fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
 432                 else
 433                         BUG();
 434                 break;
 435         case Opt_source:
 436                 if (fc->source)
 437                         return invalfc(fc, "Multiple sources specified");
 438                 return ceph_parse_source(param, fc);
 439         case Opt_mon_addr:
 440                 return ceph_parse_mon_addr(param, fc);
 441         case Opt_wsize:
 442                 if (result.uint_32 < PAGE_SIZE ||
 443                     result.uint_32 > CEPH_MAX_WRITE_SIZE)
 444                         goto out_of_range;
 445                 fsopt->wsize = ALIGN(result.uint_32, PAGE_SIZE);
 446                 break;
 447         case Opt_rsize:
 448                 if (result.uint_32 < PAGE_SIZE ||
 449                     result.uint_32 > CEPH_MAX_READ_SIZE)
 450                         goto out_of_range;
 451                 fsopt->rsize = ALIGN(result.uint_32, PAGE_SIZE);
 452                 break;
 453         case Opt_rasize:
 454                 fsopt->rasize = ALIGN(result.uint_32, PAGE_SIZE);
 455                 break;
 456         case Opt_caps_wanted_delay_min:
 457                 if (result.uint_32 < 1)
 458                         goto out_of_range;
 459                 fsopt->caps_wanted_delay_min = result.uint_32;
 460                 break;
 461         case Opt_caps_wanted_delay_max:
 462                 if (result.uint_32 < 1)
 463                         goto out_of_range;
 464                 fsopt->caps_wanted_delay_max = result.uint_32;
 465                 break;
 466         case Opt_caps_max:
 467                 if (result.int_32 < 0)
 468                         goto out_of_range;
 469                 fsopt->caps_max = result.int_32;
 470                 break;
 471         case Opt_readdir_max_entries:
 472                 if (result.uint_32 < 1)
 473                         goto out_of_range;
 474                 fsopt->max_readdir = result.uint_32;
 475                 break;
 476         case Opt_readdir_max_bytes:
 477                 if (result.uint_32 < PAGE_SIZE && result.uint_32 != 0)
 478                         goto out_of_range;
 479                 fsopt->max_readdir_bytes = result.uint_32;
 480                 break;
 481         case Opt_congestion_kb:
 482                 if (result.uint_32 < 1024) /* at least 1M */
 483                         goto out_of_range;
 484                 fsopt->congestion_kb = result.uint_32;
 485                 break;
 486         case Opt_dirstat:
 487                 if (!result.negated)
 488                         fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
 489                 else
 490                         fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
 491                 break;
 492         case Opt_rbytes:
 493                 if (!result.negated)
 494                         fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
 495                 else
 496                         fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
 497                 break;
 498         case Opt_asyncreaddir:
 499                 if (!result.negated)
 500                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
 501                 else
 502                         fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
 503                 break;
 504         case Opt_dcache:
 505                 if (!result.negated)
 506                         fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
 507                 else
 508                         fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
 509                 break;
 510         case Opt_ino32:
 511                 if (!result.negated)
 512                         fsopt->flags |= CEPH_MOUNT_OPT_INO32;
 513                 else
 514                         fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
 515                 break;
 516
 517         case Opt_fscache:
 518 #ifdef CONFIG_CEPH_FSCACHE
 519                 kfree(fsopt->fscache_uniq);
 520                 fsopt->fscache_uniq = NULL;
 521                 if (result.negated) {
 522                         fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
 523                 } else {
 524                         fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
 525                         fsopt->fscache_uniq = param->string;
 526                         param->string = NULL;
 527                 }
 528                 break;
 529 #else
 530                 return invalfc(fc, "fscache support is disabled");
 531 #endif
 532         case Opt_poolperm:
 533                 if (!result.negated)
 534                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
 535                 else
 536                         fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
 537                 break;
 538         case Opt_require_active_mds:
 539                 if (!result.negated)
 540                         fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
 541                 else
 542                         fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
 543                 break;
 544         case Opt_quotadf:
 545                 if (!result.negated)
 546                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
 547                 else
 548                         fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
 549                 break;
 550         case Opt_copyfrom:
 551                 if (!result.negated)
 552                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
 553                 else
 554                         fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
 555                 break;
 556         case Opt_acl:
 557                 if (!result.negated) {
 558 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 559                         fc->sb_flags |= SB_POSIXACL;
 560 #else
 561                         return invalfc(fc, "POSIX ACL support is disabled");
 562 #endif
 563                 } else {
 564                         fc->sb_flags &= ~SB_POSIXACL;
 565                 }
 566                 break;
 567         case Opt_wsync:
 568                 if (!result.negated)
 569                         fsopt->flags &= ~CEPH_MOUNT_OPT_ASYNC_DIROPS;
 570                 else
 571                         fsopt->flags |= CEPH_MOUNT_OPT_ASYNC_DIROPS;
 572                 break;
 573         case Opt_pagecache:
 574                 if (result.negated)
 575                         fsopt->flags |= CEPH_MOUNT_OPT_NOPAGECACHE;
 576                 else
 577                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOPAGECACHE;
 578                 break;
 579         default:
 580                 BUG();
 581         }
 582         return 0;
 583
 584 out_of_range:
 585         return invalfc(fc, "%s out of range", param->key);
 586 }
 587
 588 static void destroy_mount_options(struct ceph_mount_options *args)
 589 {
 590         dout("destroy_mount_options %p\n", args);
 591         if (!args)
 592                 return;
 593
 594         kfree(args->snapdir_name);
 595         kfree(args->mds_namespace);
 596         kfree(args->server_path);
 597         kfree(args->fscache_uniq);
 598         kfree(args->mon_addr);
 599         kfree(args);
 600 }
 601
 602 static int strcmp_null(const char *s1, const char *s2)
 603 {
 604         if (!s1 && !s2)
 605                 return 0;
 606         if (s1 && !s2)
 607                 return -1;
 608         if (!s1 && s2)
 609                 return 1;
 610         return strcmp(s1, s2);
 611 }
 612
 613 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
 614                                  struct ceph_options *new_opt,
 615                                  struct ceph_fs_client *fsc)
 616 {
 617         struct ceph_mount_options *fsopt1 = new_fsopt;
 618         struct ceph_mount_options *fsopt2 = fsc->mount_options;
 619         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
 620         int ret;
 621
 622         ret = memcmp(fsopt1, fsopt2, ofs);
 623         if (ret)
 624                 return ret;
 625
 626         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
 627         if (ret)
 628                 return ret;
 629
 630         ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
 631         if (ret)
 632                 return ret;
 633
 634         ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
 635         if (ret)
 636                 return ret;
 637
 638         ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
 639         if (ret)
 640                 return ret;
 641
 642         ret = strcmp_null(fsopt1->mon_addr, fsopt2->mon_addr);
 643         if (ret)
 644                 return ret;
 645
 646         return ceph_compare_options(new_opt, fsc->client);
 647 }
 648
 649 /**
 650  * ceph_show_options - Show mount options in /proc/mounts
 651  * @m: seq_file to write to
 652  * @root: root of that (sub)tree
 653  */
 654 static int ceph_show_options(struct seq_file *m, struct dentry *root)
 655 {
 656         struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
 657         struct ceph_mount_options *fsopt = fsc->mount_options;
 658         size_t pos;
 659         int ret;
 660
 661         /* a comma between MNT/MS and client options */
 662         seq_putc(m, ',');
 663         pos = m->count;
 664
 665         ret = ceph_print_client_options(m, fsc->client, false);
 666         if (ret)
 667                 return ret;
 668
 669         /* retract our comma if no client options */
 670         if (m->count == pos)
 671                 m->count--;
 672
 673         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
 674                 seq_puts(m, ",dirstat");
 675         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
 676                 seq_puts(m, ",rbytes");
 677         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
 678                 seq_puts(m, ",noasyncreaddir");
 679         if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
 680                 seq_puts(m, ",nodcache");
 681         if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
 682                 seq_puts(m, ",ino32");
 683         if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
 684                 seq_show_option(m, "fsc", fsopt->fscache_uniq);
 685         }
 686         if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
 687                 seq_puts(m, ",nopoolperm");
 688         if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
 689                 seq_puts(m, ",noquotadf");
 690
 691 #ifdef CONFIG_CEPH_FS_POSIX_ACL
 692         if (root->d_sb->s_flags & SB_POSIXACL)
 693                 seq_puts(m, ",acl");
 694         else
 695                 seq_puts(m, ",noacl");
 696 #endif
 697
 698         if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
 699                 seq_puts(m, ",copyfrom");
 700
 701         /* dump mds_namespace when old device syntax is in use */
 702         if (fsopt->mds_namespace && !fsopt->new_dev_syntax)
 703                 seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
 704
 705         if (fsopt->mon_addr)
 706                 seq_printf(m, ",mon_addr=%s", fsopt->mon_addr);
 707
 708         if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
 709                 seq_show_option(m, "recover_session", "clean");
 710
 711         if (!(fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS))
 712                 seq_puts(m, ",wsync");
 713
 714         if (fsopt->flags & CEPH_MOUNT_OPT_NOPAGECACHE)
 715                 seq_puts(m, ",nopagecache");
 716
 717         if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
 718                 seq_printf(m, ",wsize=%u", fsopt->wsize);
 719         if (fsopt->rsize != CEPH_MAX_READ_SIZE)
 720                 seq_printf(m, ",rsize=%u", fsopt->rsize);
 721         if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
 722                 seq_printf(m, ",rasize=%u", fsopt->rasize);
 723         if (fsopt->congestion_kb != default_congestion_kb())
 724                 seq_printf(m, ",write_congestion_kb=%u", fsopt->congestion_kb);
 725         if (fsopt->caps_max)
 726                 seq_printf(m, ",caps_max=%d", fsopt->caps_max);
 727         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
 728                 seq_printf(m, ",caps_wanted_delay_min=%u",
 729                          fsopt->caps_wanted_delay_min);
 730         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
 731                 seq_printf(m, ",caps_wanted_delay_max=%u",
 732                            fsopt->caps_wanted_delay_max);
 733         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
 734                 seq_printf(m, ",readdir_max_entries=%u", fsopt->max_readdir);
 735         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
 736                 seq_printf(m, ",readdir_max_bytes=%u", fsopt->max_readdir_bytes);
 737         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
 738                 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
 739
 740         return 0;
 741 }
 742
 743 /*
 744  * handle any mon messages the standard library doesn't understand.
 745  * return error if we don't either.
 746  */
 747 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
 748 {
 749         struct ceph_fs_client *fsc = client->private;
 750         int type = le16_to_cpu(msg->hdr.type);
 751
 752         switch (type) {
 753         case CEPH_MSG_MDS_MAP:
 754                 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
 755                 return 0;
 756         case CEPH_MSG_FS_MAP_USER:
 757                 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
 758                 return 0;
 759         default:
 760                 return -1;
 761         }
 762 }
 763
 764 /*
 765  * create a new fs client
 766  *
 767  * Success or not, this function consumes @fsopt and @opt.
 768  */
 769 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
 770                                         struct ceph_options *opt)
 771 {
 772         struct ceph_fs_client *fsc;
 773         int err;
 774
 775         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
 776         if (!fsc) {
 777                 err = -ENOMEM;
 778                 goto fail;
 779         }
 780
 781         fsc->client = ceph_create_client(opt, fsc);
 782         if (IS_ERR(fsc->client)) {
 783                 err = PTR_ERR(fsc->client);
 784                 goto fail;
 785         }
 786         opt = NULL; /* fsc->client now owns this */
 787
 788         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
 789         ceph_set_opt(fsc->client, ABORT_ON_FULL);
 790
 791         if (!fsopt->mds_namespace) {
 792                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
 793                                    0, true);
 794         } else {
 795                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
 796                                    0, false);
 797         }
 798
 799         fsc->mount_options = fsopt;
 800
 801         fsc->sb = NULL;
 802         fsc->mount_state = CEPH_MOUNT_MOUNTING;
 803         fsc->filp_gen = 1;
 804         fsc->have_copy_from2 = true;
 805
 806         atomic_long_set(&fsc->writeback_count, 0);
 807         fsc->write_congested = false;
 808
 809         err = -ENOMEM;
 810         /*
 811          * The number of concurrent works can be high but they don't need
 812          * to be processed in parallel, limit concurrency.
 813          */
 814         fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
 815         if (!fsc->inode_wq)
 816                 goto fail_client;
 817         fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
 818         if (!fsc->cap_wq)
 819                 goto fail_inode_wq;
 820
 821         hash_init(fsc->async_unlink_conflict);
 822         spin_lock_init(&fsc->async_unlink_conflict_lock);
 823
 824         spin_lock(&ceph_fsc_lock);
 825         list_add_tail(&fsc->metric_wakeup, &ceph_fsc_list);
 826         spin_unlock(&ceph_fsc_lock);
 827
 828         return fsc;
 829
 830 fail_inode_wq:
 831         destroy_workqueue(fsc->inode_wq);
 832 fail_client:
 833         ceph_destroy_client(fsc->client);
 834 fail:
 835         kfree(fsc);
 836         if (opt)
 837                 ceph_destroy_options(opt);
 838         destroy_mount_options(fsopt);
 839         return ERR_PTR(err);
 840 }
 841
 842 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
 843 {
 844         flush_workqueue(fsc->inode_wq);
 845         flush_workqueue(fsc->cap_wq);
 846 }
 847
 848 static void destroy_fs_client(struct ceph_fs_client *fsc)
 849 {
 850         dout("destroy_fs_client %p\n", fsc);
 851
 852         spin_lock(&ceph_fsc_lock);
 853         list_del(&fsc->metric_wakeup);
 854         spin_unlock(&ceph_fsc_lock);
 855
 856         ceph_mdsc_destroy(fsc);
 857         destroy_workqueue(fsc->inode_wq);
 858         destroy_workqueue(fsc->cap_wq);
 859
 860         destroy_mount_options(fsc->mount_options);
 861
 862         ceph_destroy_client(fsc->client);
 863
 864         kfree(fsc);
 865         dout("destroy_fs_client %p done\n", fsc);
 866 }
 867
 868 /*
 869  * caches
 870  */
 871 struct kmem_cache *ceph_inode_cachep;
 872 struct kmem_cache *ceph_cap_cachep;
 873 struct kmem_cache *ceph_cap_snap_cachep;
 874 struct kmem_cache *ceph_cap_flush_cachep;
 875 struct kmem_cache *ceph_dentry_cachep;
 876 struct kmem_cache *ceph_file_cachep;
 877 struct kmem_cache *ceph_dir_file_cachep;
 878 struct kmem_cache *ceph_mds_request_cachep;
 879 mempool_t *ceph_wb_pagevec_pool;
 880
 881 static void ceph_inode_init_once(void *foo)
 882 {
 883         struct ceph_inode_info *ci = foo;
 884         inode_init_once(&ci->netfs.inode);
 885 }
 886
 887 static int __init init_caches(void)
 888 {
 889         int error = -ENOMEM;
 890
 891         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
 892                                       sizeof(struct ceph_inode_info),
 893                                       __alignof__(struct ceph_inode_info),
 894                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
 895                                       SLAB_ACCOUNT, ceph_inode_init_once);
 896         if (!ceph_inode_cachep)
 897                 return -ENOMEM;
 898
 899         ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
 900         if (!ceph_cap_cachep)
 901                 goto bad_cap;
 902         ceph_cap_snap_cachep = KMEM_CACHE(ceph_cap_snap, SLAB_MEM_SPREAD);
 903         if (!ceph_cap_snap_cachep)
 904                 goto bad_cap_snap;
 905         ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
 906                                            SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 907         if (!ceph_cap_flush_cachep)
 908                 goto bad_cap_flush;
 909
 910         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
 911                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
 912         if (!ceph_dentry_cachep)
 913                 goto bad_dentry;
 914
 915         ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
 916         if (!ceph_file_cachep)
 917                 goto bad_file;
 918
 919         ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
 920         if (!ceph_dir_file_cachep)
 921                 goto bad_dir_file;
 922
 923         ceph_mds_request_cachep = KMEM_CACHE(ceph_mds_request, SLAB_MEM_SPREAD);
 924         if (!ceph_mds_request_cachep)
 925                 goto bad_mds_req;
 926
 927         ceph_wb_pagevec_pool = mempool_create_kmalloc_pool(10, CEPH_MAX_WRITE_SIZE >> PAGE_SHIFT);
 928         if (!ceph_wb_pagevec_pool)
 929                 goto bad_pagevec_pool;
 930
 931         return 0;
 932
 933 bad_pagevec_pool:
 934         kmem_cache_destroy(ceph_mds_request_cachep);
 935 bad_mds_req:
 936         kmem_cache_destroy(ceph_dir_file_cachep);
 937 bad_dir_file:
 938         kmem_cache_destroy(ceph_file_cachep);
 939 bad_file:
 940         kmem_cache_destroy(ceph_dentry_cachep);
 941 bad_dentry:
 942         kmem_cache_destroy(ceph_cap_flush_cachep);
 943 bad_cap_flush:
 944         kmem_cache_destroy(ceph_cap_snap_cachep);
 945 bad_cap_snap:
 946         kmem_cache_destroy(ceph_cap_cachep);
 947 bad_cap:
 948         kmem_cache_destroy(ceph_inode_cachep);
 949         return error;
 950 }
 951
 952 static void destroy_caches(void)
 953 {
 954         /*
 955          * Make sure all delayed rcu free inodes are flushed before we
 956          * destroy cache.
 957          */
 958         rcu_barrier();
 959
 960         kmem_cache_destroy(ceph_inode_cachep);
 961         kmem_cache_destroy(ceph_cap_cachep);
 962         kmem_cache_destroy(ceph_cap_snap_cachep);
 963         kmem_cache_destroy(ceph_cap_flush_cachep);
 964         kmem_cache_destroy(ceph_dentry_cachep);
 965         kmem_cache_destroy(ceph_file_cachep);
 966         kmem_cache_destroy(ceph_dir_file_cachep);
 967         kmem_cache_destroy(ceph_mds_request_cachep);
 968         mempool_destroy(ceph_wb_pagevec_pool);
 969 }
 970
 971 static void __ceph_umount_begin(struct ceph_fs_client *fsc)
 972 {
 973         ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
 974         ceph_mdsc_force_umount(fsc->mdsc);
 975         fsc->filp_gen++; // invalidate open files
 976 }
 977
 978 /*
 979  * ceph_umount_begin - initiate forced umount.  Tear down the
 980  * mount, skipping steps that may hang while waiting for server(s).
 981  */
 982 void ceph_umount_begin(struct super_block *sb)
 983 {
 984         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
 985
 986         dout("ceph_umount_begin - starting forced umount\n");
 987         if (!fsc)
 988                 return;
 989         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
 990         __ceph_umount_begin(fsc);
 991 }
 992
 993 static const struct super_operations ceph_super_ops = {
 994         .alloc_inode    = ceph_alloc_inode,
 995         .free_inode     = ceph_free_inode,
 996         .write_inode    = ceph_write_inode,
 997         .drop_inode     = generic_delete_inode,
 998         .evict_inode    = ceph_evict_inode,
 999         .sync_fs        = ceph_sync_fs,
1000         .put_super      = ceph_put_super,
1001         .show_options   = ceph_show_options,
1002         .statfs         = ceph_statfs,
1003         .umount_begin   = ceph_umount_begin,
1004 };
1005
1006 /*
1007  * Bootstrap mount by opening the root directory.  Note the mount
1008  * @started time from caller, and time out if this takes too long.
1009  */
1010 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
1011                                        const char *path,
1012                                        unsigned long started)
1013 {
1014         struct ceph_mds_client *mdsc = fsc->mdsc;
1015         struct ceph_mds_request *req = NULL;
1016         int err;
1017         struct dentry *root;
1018
1019         /* open dir */
1020         dout("open_root_inode opening '%s'\n", path);
1021         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1022         if (IS_ERR(req))
1023                 return ERR_CAST(req);
1024         req->r_path1 = kstrdup(path, GFP_NOFS);
1025         if (!req->r_path1) {
1026                 root = ERR_PTR(-ENOMEM);
1027                 goto out;
1028         }
1029
1030         req->r_ino1.ino = CEPH_INO_ROOT;
1031         req->r_ino1.snap = CEPH_NOSNAP;
1032         req->r_started = started;
1033         req->r_timeout = fsc->client->options->mount_timeout;
1034         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
1035         req->r_num_caps = 2;
1036         err = ceph_mdsc_do_request(mdsc, NULL, req);
1037         if (err == 0) {
1038                 struct inode *inode = req->r_target_inode;
1039                 req->r_target_inode = NULL;
1040                 dout("open_root_inode success\n");
1041                 root = d_make_root(inode);
1042                 if (!root) {
1043                         root = ERR_PTR(-ENOMEM);
1044                         goto out;
1045                 }
1046                 dout("open_root_inode success, root dentry is %p\n", root);
1047         } else {
1048                 root = ERR_PTR(err);
1049         }
1050 out:
1051         ceph_mdsc_put_request(req);
1052         return root;
1053 }
1054
1055 /*
1056  * mount: join the ceph cluster, and open root directory.
1057  */
1058 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
1059                                       struct fs_context *fc)
1060 {
1061         int err;
1062         unsigned long started = jiffies;  /* note the start time */
1063         struct dentry *root;
1064
1065         dout("mount start %p\n", fsc);
1066         mutex_lock(&fsc->client->mount_mutex);
1067
1068         if (!fsc->sb->s_root) {
1069                 const char *path = fsc->mount_options->server_path ?
1070                                      fsc->mount_options->server_path + 1 : "";
1071
1072                 err = __ceph_open_session(fsc->client, started);
1073                 if (err < 0)
1074                         goto out;
1075
1076                 /* setup fscache */
1077                 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
1078                         err = ceph_fscache_register_fs(fsc, fc);
1079                         if (err < 0)
1080                                 goto out;
1081                 }
1082
1083                 dout("mount opening path '%s'\n", path);
1084
1085                 ceph_fs_debugfs_init(fsc);
1086
1087                 root = open_root_dentry(fsc, path, started);
1088                 if (IS_ERR(root)) {
1089                         err = PTR_ERR(root);
1090                         goto out;
1091                 }
1092                 fsc->sb->s_root = dget(root);
1093         } else {
1094                 root = dget(fsc->sb->s_root);
1095         }
1096
1097         fsc->mount_state = CEPH_MOUNT_MOUNTED;
1098         dout("mount success\n");
1099         mutex_unlock(&fsc->client->mount_mutex);
1100         return root;
1101
1102 out:
1103         mutex_unlock(&fsc->client->mount_mutex);
1104         return ERR_PTR(err);
1105 }
1106
1107 static int ceph_set_super(struct super_block *s, struct fs_context *fc)
1108 {
1109         struct ceph_fs_client *fsc = s->s_fs_info;
1110         int ret;
1111
1112         dout("set_super %p\n", s);
1113
1114         s->s_maxbytes = MAX_LFS_FILESIZE;
1115
1116         s->s_xattr = ceph_xattr_handlers;
1117         fsc->sb = s;
1118         fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
1119
1120         s->s_op = &ceph_super_ops;
1121         s->s_d_op = &ceph_dentry_ops;
1122         s->s_export_op = &ceph_export_ops;
1123
1124         s->s_time_gran = 1;
1125         s->s_time_min = 0;
1126         s->s_time_max = U32_MAX;
1127         s->s_flags |= SB_NODIRATIME | SB_NOATIME;
1128
1129         ret = set_anon_super_fc(s, fc);
1130         if (ret != 0)
1131                 fsc->sb = NULL;
1132         return ret;
1133 }
1134
1135 /*
1136  * share superblock if same fs AND options
1137  */
1138 static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
1139 {
1140         struct ceph_fs_client *new = fc->s_fs_info;
1141         struct ceph_mount_options *fsopt = new->mount_options;
1142         struct ceph_options *opt = new->client->options;
1143         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1144
1145         dout("ceph_compare_super %p\n", sb);
1146
1147         if (compare_mount_options(fsopt, opt, fsc)) {
1148                 dout("monitor(s)/mount options don't match\n");
1149                 return 0;
1150         }
1151         if ((opt->flags & CEPH_OPT_FSID) &&
1152             ceph_fsid_compare(&opt->fsid, &fsc->client->fsid)) {
1153                 dout("fsid doesn't match\n");
1154                 return 0;
1155         }
1156         if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
1157                 dout("flags differ\n");
1158                 return 0;
1159         }
1160
1161         if (fsc->blocklisted && !ceph_test_mount_opt(fsc, CLEANRECOVER)) {
1162                 dout("client is blocklisted (and CLEANRECOVER is not set)\n");
1163                 return 0;
1164         }
1165
1166         if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
1167                 dout("client has been forcibly unmounted\n");
1168                 return 0;
1169         }
1170
1171         return 1;
1172 }
1173
1174 /*
1175  * construct our own bdi so we can control readahead, etc.
1176  */
1177 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1178
1179 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1180 {
1181         int err;
1182
1183         err = super_setup_bdi_name(sb, "ceph-%ld",
1184                                    atomic_long_inc_return(&bdi_seq));
1185         if (err)
1186                 return err;
1187
1188         /* set ra_pages based on rasize mount option? */
1189         sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1190
1191         /* set io_pages based on max osd read size */
1192         sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1193
1194         return 0;
1195 }
1196
1197 static int ceph_get_tree(struct fs_context *fc)
1198 {
1199         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1200         struct ceph_mount_options *fsopt = pctx->opts;
1201         struct super_block *sb;
1202         struct ceph_fs_client *fsc;
1203         struct dentry *res;
1204         int (*compare_super)(struct super_block *, struct fs_context *) =
1205                 ceph_compare_super;
1206         int err;
1207
1208         dout("ceph_get_tree\n");
1209
1210         if (!fc->source)
1211                 return invalfc(fc, "No source");
1212         if (fsopt->new_dev_syntax && !fsopt->mon_addr)
1213                 return invalfc(fc, "No monitor address");
1214
1215         /* create client (which we may/may not use) */
1216         fsc = create_fs_client(pctx->opts, pctx->copts);
1217         pctx->opts = NULL;
1218         pctx->copts = NULL;
1219         if (IS_ERR(fsc)) {
1220                 err = PTR_ERR(fsc);
1221                 goto out_final;
1222         }
1223
1224         err = ceph_mdsc_init(fsc);
1225         if (err < 0)
1226                 goto out;
1227
1228         if (ceph_test_opt(fsc->client, NOSHARE))
1229                 compare_super = NULL;
1230
1231         fc->s_fs_info = fsc;
1232         sb = sget_fc(fc, compare_super, ceph_set_super);
1233         fc->s_fs_info = NULL;
1234         if (IS_ERR(sb)) {
1235                 err = PTR_ERR(sb);
1236                 goto out;
1237         }
1238
1239         if (ceph_sb_to_client(sb) != fsc) {
1240                 destroy_fs_client(fsc);
1241                 fsc = ceph_sb_to_client(sb);
1242                 dout("get_sb got existing client %p\n", fsc);
1243         } else {
1244                 dout("get_sb using new client %p\n", fsc);
1245                 err = ceph_setup_bdi(sb, fsc);
1246                 if (err < 0)
1247                         goto out_splat;
1248         }
1249
1250         res = ceph_real_mount(fsc, fc);
1251         if (IS_ERR(res)) {
1252                 err = PTR_ERR(res);
1253                 goto out_splat;
1254         }
1255         dout("root %p inode %p ino %llx.%llx\n", res,
1256              d_inode(res), ceph_vinop(d_inode(res)));
1257         fc->root = fsc->sb->s_root;
1258         return 0;
1259
1260 out_splat:
1261         if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1262                 pr_info("No mds server is up or the cluster is laggy\n");
1263                 err = -EHOSTUNREACH;
1264         }
1265
1266         ceph_mdsc_close_sessions(fsc->mdsc);
1267         deactivate_locked_super(sb);
1268         goto out_final;
1269
1270 out:
1271         destroy_fs_client(fsc);
1272 out_final:
1273         dout("ceph_get_tree fail %d\n", err);
1274         return err;
1275 }
1276
1277 static void ceph_free_fc(struct fs_context *fc)
1278 {
1279         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1280
1281         if (pctx) {
1282                 destroy_mount_options(pctx->opts);
1283                 ceph_destroy_options(pctx->copts);
1284                 kfree(pctx);
1285         }
1286 }
1287
1288 static int ceph_reconfigure_fc(struct fs_context *fc)
1289 {
1290         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1291         struct ceph_mount_options *fsopt = pctx->opts;
1292         struct ceph_fs_client *fsc = ceph_sb_to_client(fc->root->d_sb);
1293
1294         if (fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS)
1295                 ceph_set_mount_opt(fsc, ASYNC_DIROPS);
1296         else
1297                 ceph_clear_mount_opt(fsc, ASYNC_DIROPS);
1298
1299         if (strcmp_null(fsc->mount_options->mon_addr, fsopt->mon_addr)) {
1300                 kfree(fsc->mount_options->mon_addr);
1301                 fsc->mount_options->mon_addr = fsopt->mon_addr;
1302                 fsopt->mon_addr = NULL;
1303                 pr_notice("ceph: monitor addresses recorded, but not used for reconnection");
1304         }
1305
1306         sync_filesystem(fc->root->d_sb);
1307         return 0;
1308 }
1309
1310 static const struct fs_context_operations ceph_context_ops = {
1311         .free           = ceph_free_fc,
1312         .parse_param    = ceph_parse_mount_param,
1313         .get_tree       = ceph_get_tree,
1314         .reconfigure    = ceph_reconfigure_fc,
1315 };
1316
1317 /*
1318  * Set up the filesystem mount context.
1319  */
1320 static int ceph_init_fs_context(struct fs_context *fc)
1321 {
1322         struct ceph_parse_opts_ctx *pctx;
1323         struct ceph_mount_options *fsopt;
1324
1325         pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1326         if (!pctx)
1327                 return -ENOMEM;
1328
1329         pctx->copts = ceph_alloc_options();
1330         if (!pctx->copts)
1331                 goto nomem;
1332
1333         pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1334         if (!pctx->opts)
1335                 goto nomem;
1336
1337         fsopt = pctx->opts;
1338         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1339
1340         fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1341         fsopt->rsize = CEPH_MAX_READ_SIZE;
1342         fsopt->rasize = CEPH_RASIZE_DEFAULT;
1343         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1344         if (!fsopt->snapdir_name)
1345                 goto nomem;
1346
1347         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1348         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1349         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1350         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1351         fsopt->congestion_kb = default_congestion_kb();
1352
1353 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1354         fc->sb_flags |= SB_POSIXACL;
1355 #endif
1356
1357         fc->fs_private = pctx;
1358         fc->ops = &ceph_context_ops;
1359         return 0;
1360
1361 nomem:
1362         destroy_mount_options(pctx->opts);
1363         ceph_destroy_options(pctx->copts);
1364         kfree(pctx);
1365         return -ENOMEM;
1366 }
1367
1368 static void ceph_kill_sb(struct super_block *s)
1369 {
1370         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1371
1372         dout("kill_sb %p\n", s);
1373
1374         ceph_mdsc_pre_umount(fsc->mdsc);
1375         flush_fs_workqueues(fsc);
1376
1377         kill_anon_super(s);
1378
1379         fsc->client->extra_mon_dispatch = NULL;
1380         ceph_fs_debugfs_cleanup(fsc);
1381
1382         ceph_fscache_unregister_fs(fsc);
1383
1384         destroy_fs_client(fsc);
1385 }
1386
1387 static struct file_system_type ceph_fs_type = {
1388         .owner          = THIS_MODULE,
1389         .name           = "ceph",
1390         .init_fs_context = ceph_init_fs_context,
1391         .kill_sb        = ceph_kill_sb,
1392         .fs_flags       = FS_RENAME_DOES_D_MOVE,
1393 };
1394 MODULE_ALIAS_FS("ceph");
1395
1396 int ceph_force_reconnect(struct super_block *sb)
1397 {
1398         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1399         int err = 0;
1400
1401         fsc->mount_state = CEPH_MOUNT_RECOVER;
1402         __ceph_umount_begin(fsc);
1403
1404         /* Make sure all page caches get invalidated.
1405          * see remove_session_caps_cb() */
1406         flush_workqueue(fsc->inode_wq);
1407
1408         /* In case that we were blocklisted. This also reset
1409          * all mon/osd connections */
1410         ceph_reset_client_addr(fsc->client);
1411
1412         ceph_osdc_clear_abort_err(&fsc->client->osdc);
1413
1414         fsc->blocklisted = false;
1415         fsc->mount_state = CEPH_MOUNT_MOUNTED;
1416
1417         if (sb->s_root) {
1418                 err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1419                                         CEPH_STAT_CAP_INODE, true);
1420         }
1421         return err;
1422 }
1423
1424 static int __init init_ceph(void)
1425 {
1426         int ret = init_caches();
1427         if (ret)
1428                 goto out;
1429
1430         ceph_flock_init();
1431         ret = register_filesystem(&ceph_fs_type);
1432         if (ret)
1433                 goto out_caches;
1434
1435         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1436
1437         return 0;
1438
1439 out_caches:
1440         destroy_caches();
1441 out:
1442         return ret;
1443 }
1444
1445 static void __exit exit_ceph(void)
1446 {
1447         dout("exit_ceph\n");
1448         unregister_filesystem(&ceph_fs_type);
1449         destroy_caches();
1450 }
1451
1452 static int param_set_metrics(const char *val, const struct kernel_param *kp)
1453 {
1454         struct ceph_fs_client *fsc;
1455         int ret;
1456
1457         ret = param_set_bool(val, kp);
1458         if (ret) {
1459                 pr_err("Failed to parse sending metrics switch value '%s'\n",
1460                        val);
1461                 return ret;
1462         } else if (!disable_send_metrics) {
1463                 // wake up all the mds clients
1464                 spin_lock(&ceph_fsc_lock);
1465                 list_for_each_entry(fsc, &ceph_fsc_list, metric_wakeup) {
1466                         metric_schedule_delayed(&fsc->mdsc->metric);
1467                 }
1468                 spin_unlock(&ceph_fsc_lock);
1469         }
1470
1471         return 0;
1472 }
1473
1474 static const struct kernel_param_ops param_ops_metrics = {
1475         .set = param_set_metrics,
1476         .get = param_get_bool,
1477 };
1478
1479 bool disable_send_metrics = false;
1480 module_param_cb(disable_send_metrics, &param_ops_metrics, &disable_send_metrics, 0644);
1481 MODULE_PARM_DESC(disable_send_metrics, "Enable sending perf metrics to ceph cluster (default: on)");
1482
1483 /* for both v1 and v2 syntax */
1484 static bool mount_support = true;
1485 static const struct kernel_param_ops param_ops_mount_syntax = {
1486         .get = param_get_bool,
1487 };
1488 module_param_cb(mount_syntax_v1, &param_ops_mount_syntax, &mount_support, 0444);
1489 module_param_cb(mount_syntax_v2, &param_ops_mount_syntax, &mount_support, 0444);
1490
1491 module_init(init_ceph);
1492 module_exit(exit_ceph);
1493
1494 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1495 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1496 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1497 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1498 MODULE_LICENSE("GPL");